Trajectory evaluation and behavioral scoring using JAABA in a noisy system - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 220

what beter when applied to the nontraining data, perhaps suggesting that the training dataset
was a particularly difficult dataset for which to call sex.

Table 5
Evaluation of the Sex Classifier Performance on Nontraining Videos

Algorithms Accuracy Sensitivity Specificity

JAABA 0.919 0.941 0.877

GradientBoost 0.925 0.942 0.895

Logistic 0.872 0.940 0.779

lSVM 0.957 0.932 0.981

gSVM 0.854 0.848 0.859

The mean accuracy, sensitivity, and specificity scores across the videos are shown.
Behavior Annotation : The ultimate goal of trajectory analysis, and the implementation of
JAABA on tracking data, is to evaluate behaviors in a diversity of experimental manipulations.
In scoring Chasing, JAABA, GradientBoost, and logistic regression all performed well, with
accuracy above 0.85 ( Table 6 ). Even on the nontraining dataset, all the methods besides the
Gaussian SVM had accuracy greater than 0.8, but logistic regression and the linear SVM per-
formed best ( Table 7 ).

Table 6
Evaluation of the Chase Classifier Performance on Threefold Cross Validation

Algorithms Accuracy Sensitivity Specificity Precision AUC

JAABA 0.887 0.920 0.867 0.809 -

GradientBoost 0.884 0.919 0.827 0.896 0.917

Logistic 0.859 0.913 0.770 0.866 0.888

lSVM 0.844 0.813 0.895 0.927 0.861

gSVM 0.781 0.696 0.919 0.933 0.885

The accuracy, sensitivity, specificity, precision, and area under the curve scores are shown for each.